Meta launches Muse Spark Open, an 816B-parameter open-weight model that runs on Macs
Meta announced Muse Spark Open, an open-weight AI model with 816 billion parameters and a 1 million-token context window that the company says is designed to run on Macs. Meta says the model is available now via multiple APIs and that it runs best on a Mac Studio equipped with the M5 Ultra, while Engadget reported local use requires at least 1 terabyte of RAM.
AI-assisted synthesis · Based on 6 outlets' reporting · Full text read from 1 of 6 outletsHow it's made
File photoWhat happened
Key facts
- Meta announced Muse Spark Open, which it describes as an open-weight AI model designed to run on Macs.
- Meta says Muse Spark Open uses 816 billion parameters and a 1 million-token context window.
- Meta says it deployed Muse Spark Open to perform tasks in DaVinci Resolve, including reviewing and cropping footage.
- The model is available now through multiple API platforms, according to Meta.
- Engadget reported that running Muse Spark Open locally requires at least 1 terabyte of RAM and that it runs best on a Mac Studio with the M5 Ultra chip.
The brief
Meta announced Muse Spark Open, an open-weight artificial intelligence model the company says combines native multimodal capabilities with a very large context window and the capacity to run on Apple Macs.
Meta described the model as using 816 billion parameters and providing a 1 million-token context window, and the company said it has made the model available through multiple application programming interfaces.
TechCrunch reported Meta presented evaluation results it called evidence that Muse Spark Open is the "most capable" open-weight model released in the United States to date. Meta said it paired Muse Spark Open with Muse Code, its AI coding terminal, and used the model in agentic workflows that Meta described as producing end-to-end coding outputs.
TechCrunch wrote that Meta deployed the model to implement coding fixes for a Mastodon client, including bug fixes, new features and a full build, describing those workflows as demonstrations of the model's capabilities.
TechCrunch also said Meta applied Muse Spark Open inside DaVinci Resolve, the professional video-editing software, reporting that the model was able to review footage and perform cropping tasks. Meta presented those uses as examples of the model’s multimodal application across coding and video workflows.
Running Muse Spark Open locally, however, requires substantial hardware, according to TechCrunch's account of other reporting: Engadget reported the model needs at least 1 terabyte of RAM for local execution and runs best on a Mac Studio equipped with Apple’s M5 Ultra chip.
TechCrunch said Meta recommends Mac Studio with the M5 Ultra as the optimal hardware for on-device use, while also offering the model via APIs for remote access. Meta said the model is available now through multiple APIs, and TechCrunch framed the release as immediate availability for developers and organisations that can access the announced endpoints.
TechCrunch also noted the company’s claims about the model’s relative performance, citing Meta’s released evaluation results as the basis for calling Muse Spark Open highly capable among open-weight models in the United States.
Outlets that published headlines about the launch include Firstpost, Unite.AI and Crypto Briefing, and Meta posted a formal introduction to Muse Spark Open on its research site. The public technical and availability claims in TechCrunch’s report rest on Meta’s announcements and on Engadget’s hardware reporting about local system requirements.
Why it matters
- Mac Studio users and organisations requiring large-context multimodal models face concrete hardware costs because Engadget reported local runs of Muse Spark Open require at least 1 terabyte of RAM and Meta says it runs best on Mac Studio with the M5 Ultra.
- Developers who need an open-weight model with a long context can access Muse Spark Open now through multiple APIs, per Meta; that availability affects teams that build agentic coding workflows, as Meta demonstrated with Muse Code and the Mastodon client fixes.
- Video production teams that use DaVinci Resolve bear a specific mechanism of change: Meta says Muse Spark Open can review and crop footage, which directly alters post-production tooling for editors who adopt the model.
What to watch next
- Whether more detailed system requirements or optimization guides for Muse Spark Open are published by Meta or Engadget to clarify local deployment needs.
- Whether additional API platforms beyond those Meta announced list Muse Spark Open as available and what pricing or access terms they set.
- Whether Meta or third parties publish empirical benchmarks comparing Muse Spark Open’s performance to other open-weight models, expanding on the evaluation results TechCrunch cited.
How it's being covered
Coverage analysisShow details6 source articles · 6 unique outletsCoverage mix: Center Only· 4 CenterWhere the outlets are based: US/Canada 50% · Asia 33% · Other 17%
- Coverage is currently concentrated among Center-rated outlets (4 of 4 rated sources); comparatively few Left or Right-rated outlets have published on the event.
How outlets' accounts compare
How we got here
Meta’s immediate backdrop was its public release of the LLaMA 2 family on June 1, 2023, when the company made large-language-model weights and licenses broadly available to researchers and some commercial partners. That followed Meta’s initial LLaMA announcement on February 21, 2023, which introduced a family of transformer-based models as alternatives to closed models.
Before LLaMA, Meta released Code Llama on August 31, 2023, targeting coding tasks with specialized fine-tuning. The structural roots for open-weight models trace to the transformer architecture introduced by Vaswani et al. in 2017 and to the broader industry shift toward large-parameter models after OpenAI’s GPT-3 (175 billion parameters) in 2020.
Original sources
6 articles from 6 outletsEvery story links to the reporting it was built from. How we choose sources
Spotted a factual error, a broken link or a misleading summary? Tell us.